What can siblings and cousins tell us about the role of early life family income in the aetiologies of violent crime, substance misuse and psychiatric morbidity?

In our recent paper,¹ we examined associations between family income during childhood and adolescence on subsequent risks of developing psychiatric disorders, having substance misuse problems and being arrested for a violent crime. We used nationwide population data on 650 680 individuals born in Finland between 1986 and 1996 and their siblings. Our initial analyses indicated, as expected on a population level, that higher family income was associated with lower risks of all examined outcomes. We then used a sibling-comparison design, where we compared risks of the outcomes between biological full-siblings who, owing to the fact that they were born during different years, grew up in the same households during periods when the parents had varying levels of income. If the reported associations were consistent with a causal inference, we would have expected the siblings who were exposed to lower family income relative to their co-siblings to have higher risks of the outcomes, but we did not find any support for this. In fact, the point estimates for the sibling-comparison estimates were all close to 1, indicating no difference between the siblings regardless of their family income exposure, and the confidence intervals were narrow, indicating a high level of precision of the estimates. Complementary sensitivity analyses indicated no moderation effects by demographic characteristics (e.g. sex, birth year and birth order) and stability of the findings across a wide range of alternative exposure and outcome definitions. In response to the points raised in the letter Rod et al.,² we have the following comments.

First, Rod et al. question whether the sibling design used can ‘inform the causal question that is being raised (and supposedly answered), namely whether there will be an effect of intervening on family income’. In our view, this misrepresents the implications of the findings that we discussed. We explicitly stated that we find it unlikely (but not impossible), on the basis of our findings, that interventions that focus on increasing parental income levels will lead to lower offspring outcome rates if the findings are replicated. In fact, we recommended large-scale experimental studies of interventions directly aimed at alleviating childhood poverty to be conducted with long follow-ups to clarify whether they have an impact on adult outcomes. The challenge with such studies is to design them in a way that is both ethically justifiable and practically feasible. In addition, we stated that increasing parental income may be beneficial for outcomes that we had not examined.

Second, Rod et al. question to what extent the exposure-discordant siblings were representative of the general population of Finland. It is important to bear in mind that there are no direct ways of testing for the external validity of sibling-comparison estimates as this would require population-wide causal effects to be identified and being used as reference points. If this was possible to do, it would render the sibling-comparisons a relatively pointless exercise. Instead, there are alternative approaches to indirectly test for the representativeness. We adopted two such approaches. In the first approach, we compared the population-wide associations between family income and outcomes in the full sample with a subset that included all siblings. We found that these estimates were commensurate with one another, suggesting the absence of systematic differences across participants. The second approach was cousin comparisons, where we compared outcome risks in differentially exposed cousins. Given that cousins are less similar in their parental exposures than siblings, a larger number of differentially exposed cousins is observed than siblings in a given population sample. Consistent findings using cousin-comparisons thus strengthens the external validity of the sibling-comparisons.³ In our study, we found that the cousin-comparison estimates were attenuated by approximately 40% across the outcomes, despite accounting for 12.5% of co-segregating genes. Shared environmental influences among cousins are rare in the Nordic context, and in complementary sensitivity analyses, we additionally tested for this by only including cousins who resided in different municipalities. We found negligible differences in between these different cousin estimates.

This point is also relevant to the criticism raised by Ledberg et al.,⁴ namely that our sibling-comparison models may have over-adjusted mediators, such as the neighbourhood residence of the participants. If neighborhoods and related measures (i.e. schools) had any explanatory power in our sample, we would have expected to estimate larger associations in cousin pairs who resided in different municipalities compared with all cousin pairs, which we did not find. Importantly, we have previously reported that Swedish siblings who grew up in families that moved between neighbourhoods with varying socioeconomic status did not significantly differ from one another in their subsequent rates of antisocial behaviours and psychiatric morbidity.^5,6 Such analyses imply that area effects are modest at best.

Third, Rod et al. argue that our findings could potentially be explained by carry-over effects. They provided an example of a family in which an older sibling developed a psychiatric disorder, which in turn caused a reduction of the family income exposure of their younger sibling. The cousin-comparison design is also used to indirectly test for such carry-over effects.⁷ To take the same example, the psychiatric disorder of the older sibling would not have affected the exposure of the family income exposure of their cousins. Despite this, we found clear reductions of population-wide associations when comparing differentially exposed cousins. We further note that we did not find any support for within-family moderation effects by either birth year or birth order, which could have potentially been suggestive of carry-over effects.

Fourth, we estimated the within-family variation in the income measures using sibling intraclass correlations. We found that up to a third of the variance in the family income measures were not shared by siblings growing up in the same household, which explains why we were able to estimate the sibling-comparison associations with a relatively large degree of precision.

One final concern, that both letters raise, is that studies based on sibling designs lead to strong social policy implications, something which we did not formulate in our paper. We note, however, that policy recommendations based on observational data without due consideration for unmeasured confounding is likely to lead to exaggerated hopes of policy effectiveness.

Author contributions

AS and SF conceived of and drafted the letter. All authors critically revised the draft, and approved the final version.

Funding

The letter was supported by the Wellcome Trust (#202836/Z/16/Z) and the Academy of Finland (#308247, #294861, and #316595). AS is supported by the National Institute for Health Research (NIHR) Oxford Health Biomedical Research Centre (BRC). The funders were not involved in the preparation, review or approval of the letter.

Conflict of interest

None declared.